Optimal prediction of PKS: RSO modified alkyd resin polycondensation process using discrete-delayed observations, ANN and RSM-GA techniques

Abstract

Alkyd resins are widely used in the paint industry and although they have a long history about 70–100 years, today the developments in alkyds are still welcome and innovations are still needed. Artificial neural network (ANN) and response surface methodology based on a 25−1 fractional factorial design were used as tools for simulation and optimization of the polycondensation process for autooxidative drying alkyd resin from palm kernel stearin: rubber seed oil blend of 70:30 ratio. A feed forward neural network model with Levenberg–Marquardt back propagation training algorithm was adapted to predict the responses (conversion Y 1, viscosity Y 2, and molecular weight average Y 3). The studied input variables were reaction time, temperature, catalyst concentration, oil ratio, and stirring rate. The performance of the RSM and ANN model showed adequate prediction of the responses in terms of the process factors, with MRPD of ±4.47% (Y 1), ±2.08% (Y 2), ±8.92% (Y 3) and ±6.50% (Y 1), ±3.31% (Y 2), ±10.20% (Y 3), respectively. The sensitivity analysis showed that while reaction time is the most effective process parameter, the interaction of the five process variables produced the most significant effect on the studied responses with the overall minimum MSE of 0.079. The optimization task performed using a genetic algorithm linked to the RSM model gave a viable, nondominated optimal response and optimum operating conditions regarding the route to high-quality resin at reduced material and operational costs. Overall, coupled RSM-GA was found to be a better tool for modeling and optimization of the alkyd resin production.